Learning device, estimation device, learning method, estimation method, and program

Abstract

To reduce a burden required for determining a cost of each position in a route planning problem.SOLUTION: A learning device comprises a control unit that performs learning of a learning target model that is a mathematical model of a learning target, where the learning target model is a mathematical model that estimates, based on first input information indicating information regarding each position in a two dimensional or three dimensional space and second input information that is text data indicating, in a natural language, a condition satisfied by a movement path of a moving object that moves in the space, a cost of each position in a path planning problem of obtaining a movement path in the space of the moving object that satisfies the condition as an optimal solution.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a learning device, an estimation device, a learning method, an estimation method, and a program. [Background technology]

[0002] There is technology to set and solve path planning problems, which are problems that involve planning a route from a start to a goal in a certain environment with the lowest cumulative cost, such as route search in car navigation systems and map apps, navigation for the autonomous movement of mobile robots, and game AI. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Car navigation route search technology (2004 Patent Distribution Support Chart) [0nline], [Retrieved May 24, 2024], Internet<URL:https: / / www.inpit.go.jp / blob / katsuyo / pdf / chart / fdenki22.pdf> Summary of the Invention [Problem to be solved by the invention]

[0004] The optimal solution to a path planning problem varies depending on the cost assigned to each location. For example, if a binary value representing whether or not a moving object has passed through each location is assigned to each location, the optimal solution is the shortest route. For example, if a value representing the degree of road roughness at each location is assigned to each location, the optimal solution is the route with the least roughness.

[0005] In this way, the optimal solution to a path planning problem varies depending on the cost set for each position. Therefore, in order to find a path that satisfies desired conditions using a path planning problem, it is necessary to appropriately set the cost for each position before solving the path planning problem. However, setting the cost for each position is not always easy and can be a heavy burden, for example, for those who are not familiar with the design and implementation of path planning problems.

[0006] In view of the above circumstances, an object of the present invention is to provide a technique that reduces the burden required to determine the cost of each position in a path planning problem. [Means for solving the problem]

[0007] One aspect of the present invention is a learning device that includes a control unit that learns a learning object model, which is a mathematical model of a learning object, and the learning object model is a mathematical model that estimates the cost of each position in a path planning problem that obtains a path of movement within a space of a moving object that satisfies the conditions as an optimal solution, based on first input information that indicates information about each position in a two- or three-dimensional space and second input information that is text data in natural language indicating conditions that must be satisfied by the path of movement of the moving object moving through the space.

[0008] One aspect of the present invention is an estimation device comprising: a control unit that learns a learning object model, which is a mathematical model of a learning object, and the learning object model is a mathematical model that estimates the cost of each position in a two- or three-dimensional space in a path planning problem that obtains a movement path within the space of a moving object that satisfies the conditions as an optimal solution, based on first input information that indicates information about each position in the space and second input information that is text data in natural language indicating conditions that must be satisfied by the movement path of the moving object moving through the space, and an estimation unit that performs estimation using the learned learning object model obtained by the learning device.

[0009] One aspect of the present invention is a learning method including a control step in which a computer learns a learning object model, which is a mathematical model of a learning object, and the learning object model is a mathematical model that estimates the cost of each position in a path planning problem in which a path of movement of a moving object within the space that satisfies the conditions as an optimal solution is obtained based on first input information indicating information about each position in a two- or three-dimensional space and second input information being text data indicating, in natural language, conditions to be satisfied by the path of movement of the moving object moving through the space.

[0010] One aspect of the present invention is an estimation method including a control step in which a computer learns a learning object model, which is a mathematical model of a learning object, and the learning object model is a mathematical model that estimates the cost of each position in a two- or three-dimensional space in a path planning problem that obtains a movement path within the space of a moving object that satisfies the conditions as an optimal solution, based on first input information that indicates information about each position in the space and second input information that is text data in natural language that indicates conditions that must be satisfied by the movement path of the moving object moving through the space, and an estimation step in which a computer performs estimation using the learned learning object model obtained by the learning method.

[0011] One aspect of the present invention is a program for causing a computer to function as the learning device described above.

[0012] One aspect of the present invention is a program for causing a computer to function as the above-described estimation device. [Effects of the Invention]

[0013] The present invention allows for a reduction in the burden of determining the cost of each position in a path planning problem. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 1 is an explanatory diagram illustrating an information processing system according to an embodiment. [Figure 2] FIG. 2 is a diagram showing a first example of a movement route in the embodiment. [Figure 3]FIG. 10 is a diagram showing a second example of a movement path in the embodiment. [Figure 4] FIG. 10 is a diagram showing a third example of a movement path in the embodiment. [Figure 5] FIG. 2 is a diagram illustrating an example of a hardware configuration of a learning device according to an embodiment. [Figure 6] 10 is a flowchart showing an example of a flow of processing executed by a learning device according to an embodiment. [Figure 7] FIG. 2 is a diagram illustrating an example of a hardware configuration of an estimation apparatus according to an embodiment. [Figure 8] 1 is a flowchart showing an example of a flow of processing executed by an estimation device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0015] (Embodiment) 1 is an explanatory diagram illustrating an information processing system 100 according to an embodiment. The information processing system 100 includes a learning device 1 and an estimation device 2.

[0016] The learning device 1 includes a control unit 11 having a processor 91, such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an NPU (Neural Network Processing Unit), and a memory 92, which are connected via a bus, and executes a program.

[0017] The control unit 11 executes, for example, a learning process. The learning process is a process for learning a learning target model, which is a mathematical model of a learning target. The learning target model is a mathematical model that estimates a position cost, which is the cost of each position in a path planning problem.

[0018] More specifically, the learning target model is a mathematical model that estimates a location cost based on first input information and second input information. The first input information indicates information about each location in a two- or three-dimensional space. Hereinafter, the two- or three-dimensional space in which the first input information indicates information about each location is referred to as a target space. The information about each location in the target space is represented, for example, by a feature vector.

[0019] The second input information is text data in natural language indicating conditions (hereinafter referred to as "path conditions") to be satisfied by the path of a moving object moving through the space (i.e., the target space) in which the first input information indicates information about each position. The path planning problem for which the trained model estimates the position cost is, more specifically, a path planning problem for obtaining, as an optimal solution, a path within the space that satisfies the path conditions.

[0020] Learning is performed until a predetermined condition for terminating learning (hereinafter referred to as "learning termination condition") is satisfied. The learning termination condition may be any condition related to the termination of learning, such as a condition that the model to be trained has been updated a predetermined number of times, or a condition that the change in the model to be trained due to the update is smaller than a predetermined change. The model to be trained at the time the learning termination condition is satisfied is the trained model.

[0021] The learning may be supervised learning or unsupervised learning. In the case of supervised learning, data to which a correct label is assigned for each pair of first input information and second input information is used as training data. The correct label indicates, for example, the correct location cost. In other words, the correct label indicates the true cost of the location cost according to the travel route conditions indicated by the second input information. The correct label may indicate, for example, the correct travel route. In this case, in other words, the correct label indicates the true travel route according to the travel route conditions indicated by the second input information.

[0022] In the route optimization problem, an optimal solution is obtained based on the accumulation of location costs (hereinafter referred to as "accumulated cost"). The route optimization problem may be a problem in which a travel route that minimizes the accumulated cost is obtained as the optimal solution travel route, or a problem in which a travel route that maximizes the accumulated cost is obtained as the optimal solution travel route.

[0023] When the training model estimates the location cost of each location, the training process updates the training model to reduce the difference between the location cost of each location estimated by the training model and the true cost of each location indicated by the correct label.When the training model estimates the movement path of a moving object, the training process updates the training model to reduce the difference between the movement path estimated by the training model and the true movement path indicated by the correct label.

[0024] The learning target model may acquire an embedding vector of the second input information based on the second input information, which is text data, and perform estimation using the acquired embedding vector.

[0025] <Example of first input information> The first input information may be the result of a predetermined embedding process for information indicating the target space, such as a photograph of the target space. The predetermined embedding process may be, for example, image segmentation. Information about each position indicated by the first input information (hereinafter referred to as "positional relationship information") may indicate, for example, the classification to which each position belongs. For example, when the predetermined embedding process is image segmentation, the positional relationship information may indicate the classification to which each position belongs. Therefore, the first input information may be, for example, information indicating an occupancy grid map. The positional relationship information may indicate, for example, the degree of unevenness at each position, or the physical characteristics of each position, such as the shape or height of an obstacle at each position.

[0026] <Example of second input information> An example of second input information is shown below. As described above, the second input information is text data indicating travel route conditions in natural language. Therefore, the second input information may be, for example, text data of text indicating physical characteristics of the travel route in natural language, such as "take wide roads" or "avoid uneven roads." The second input information may also be, for example, text data of text indicating, in natural language, passing through locations with predetermined tangible or intangible value, such as "pass through tourist attractions."

[0027] The second input information may also be text data that indicates in natural language the task to be given to the mobile object, such as "Carry the food carefully." The second input information may also be text data that indicates the attributes of the mobile object, such as "Move the robot with small wheels" or "I want to move around in a wheelchair." The text "Move the robot with small wheels" indicates that the mobile object is a robot with small wheels. The text "I want to move around in a wheelchair" indicates that the mobile object is a wheelchair.

[0028] In the example of Figure 1, the second input information is text data indicating the text "Find a wide road," and the first input information in the example of Figure 1 indicates map M1. A map is information indicating what each location in space is like. Map M1 in the example of Figure 1 indicates whether each location is a wall or an aisle. Therefore, in the example of Figure 1, the trained model estimates the location cost to be assigned to each location in the space indicated by map M1 so as to obtain a route optimization problem in which the narrowest road among multiple routes in the space indicated by map M1 is obtained as the optimal solution. Image M2 is an example of the estimation result of the trained model.

[0029] FIG. 2 is a diagram showing a first example of a travel route in an embodiment. FIG. 2 shows a travel route K1 as an example of a travel route existing in the space shown by the map M1 in FIG. 1. FIG. 3 is a diagram showing a second example of a travel route in an embodiment. FIG. 3 shows a travel route K2 as an example of a travel route existing in the space shown by the map M1 in FIG. 1. FIG. 4 is a diagram showing a second example of a travel route in an embodiment. FIG. 4 shows a travel route K3 as an example of a travel route existing in the space shown by the map M1 in FIG. Of the travel routes K1 to K3, the travel route K2 is an example of a travel route that follows the text indicated by the second input information exemplified in FIG. 1.

[0030] 2 to 4, multiple black circles are shown on the movement routes K2 to K4. The position cost is, for example, a cost determined at the position of each black circle. In this case, the cumulative cost of the movement route K2 is, for example, the sum of the costs at the positions of each black circle on the movement route K2. The cumulative cost of the movement route K3 is, for example, the sum of the costs at the positions of each black circle on the movement route K3. The cumulative cost of the movement route K4 is, for example, the sum of the costs at the positions of each black circle on the movement route K4.

[0031] <Effects of learning processing> As described above, the learning target model is a mathematical model that estimates the location cost of each location in a route planning problem that obtains a movement path of a moving object that satisfies the movement path conditions indicated by the second input information as an optimal solution, based on the first input information and the second input information. Therefore, it can be said that the execution of the learning process improves the estimation accuracy of the mathematical model that estimates the location cost based on text expressed in natural language. Because this mathematical model estimates the location cost, it can be said that the learning process improves the estimation accuracy of the location cost. Therefore, the learning device 1 that executes the learning process can reduce the burden required to determine the cost of each location in the route planning problem.

[0032] The training model also estimates location costs based on second input information. The second input information is text data in which travel route conditions are written in natural language. Therefore, a user using the trained training model can specify travel route conditions in natural language. Because instructions are in natural language, even someone without computer programming knowledge can have the trained training model estimate location costs. Therefore, the mathematical model obtained by the training process can reduce the burden on someone without programming skills to learn programming skills in order to design a travel route. Therefore, in this sense, the training device 1 that executes the training process can reduce the burden required to determine the cost of each location in a route planning problem.

[0033] <Estimation device 2> The estimation device 2 performs an estimation process. The estimation process is an estimation process using a trained learning object model. More specifically, the estimation process is a process of estimating a location cost at each position in a space indicated by the first input information, based on first input information and second input information, using the trained learning object model. Therefore, the estimation device 2, for example, receives input of the first input information and the second input information, and estimates a location cost at each position in a space indicated by the first input information, using the trained learning object model.

[0034] <Example of hardware configuration of learning device 1> 5 is a diagram showing an example of the hardware configuration of the learning device 1 according to the embodiment. The learning device 1 includes a control unit 11 and executes a program, and functions as a device including the control unit 11, an interface unit 12 including a communication interface 121, and a storage unit 13 by executing the program.

[0035] More specifically, the processor 91 reads out a program stored in the storage unit 13 and stores the read out program in the memory 92. When the processor 91 executes the program stored in the memory 92, the learning device 1 functions as a device including the control unit 11, the interface unit 12, and the storage unit 13.

[0036] The control unit 11 controls the operation of each functional unit included in the learning device 1. The control unit 11 executes, for example, the learning process as described above. The control unit 11 acquires, for example, information stored in the memory unit 13. Specifically, the process of acquiring the information stored in the memory unit 13 is reading.

[0037] The interface unit 12 includes a communication interface for connecting the learning device 1 to an external device. The interface unit 12 communicates with the external device via wired or wireless communication.

[0038] The external device is, for example, a device that transmits information used for learning, including first input information and second input information. The information used for learning may include, for example, a correct answer label. The interface unit 12 acquires the information used for learning by communicating with the device that transmits the information used for learning.

[0039] The external device may be, for example, the estimation device 2. In this case, the estimation device 2 can execute the trained learning object model obtained by executing the learning process through communication via the interface unit 12.

[0040] Interface unit 12 may be configured to include input devices such as a mouse, keyboard, or touch panel. Interface unit 12 may be configured as an interface that connects these input devices to learning device 1. In this way, the input devices of interface unit 12 accept input of various information to learning device 1 via wired or wireless connections. Note that information does not necessarily have to be input to the communication interface of interface unit 12, but may also be input to the input devices of interface unit 12.

[0041] Interface unit 12 outputs, for example, various types of information. Interface unit 12 is configured to include a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display, and a speaker. Interface unit 12 may be configured as an interface that connects these display devices or speakers to learning device 1. Therefore, interface unit 12 may output, for example, information input to an input device of interface unit 12 as an image or sound.

[0042] The storage unit 13 is configured using a computer-readable storage medium (non-transitory computer-readable recording medium) such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 13 stores various information related to the learning device 1. The storage unit 13 stores various information generated by the operation of the control unit 11, for example. The storage unit 13 may exist on a cloud, for example.

[0043] 6 is a flowchart showing an example of the flow of processing executed by the learning device 1 in the embodiment. The control unit 11 executes the learning processing (step S101).

[0044] <Example of hardware configuration of estimation device 2> 7 is a diagram illustrating an example of a hardware configuration of the estimation device 2 according to an embodiment. The estimation device 2 includes a control unit 21 including a processor 93 such as a CPU, GPU, or NPU, and a memory 94, which are connected via a bus, and executes a program. By executing the program, the estimation device 2 functions as a device including the control unit 21, an interface unit 22, and a storage unit 23.

[0045] More specifically, the processor 93 reads out a program stored in the storage unit 23 and stores the read program in the memory 94. The processor 93 executes the program stored in the memory 94, causing the estimation device 2 to function as a device including the control unit 21, the interface unit 22, and the storage unit 23.

[0046] The control unit 21 controls the operation of each functional unit included in the estimation device 2. The control unit 21 executes, for example, an estimation process. The control unit 21 acquires, for example, information stored in the memory unit 23. Specifically, the process of acquiring the information stored in the memory unit 23 is reading.

[0047] The interface unit 22 includes a communication interface for connecting the estimation device 2 to an external device. The interface unit 22 communicates with the external device via wire or wirelessly.

[0048] The external device is, for example, a device that transmits first input information and second input information used to estimate location costs using a trained learning object model. The interface unit 22 acquires the first input information and the second input information by communicating with the device that transmits the first input information and the second input information.

[0049] The external device may be, for example, the learning device 1. In this case, the estimation device 2 can execute the learned learning object model obtained by the learning device 1 through communication via the interface unit 22.

[0050] The interface unit 22 may be configured to include input devices such as a mouse, a keyboard, a touch panel, etc. The interface unit 22 may be configured as an interface that connects these input devices to the estimation device 2. In this way, the input devices of the interface unit 22 accept input of various information to the estimation device 2 via wired or wireless connections. Note that information does not necessarily have to be input to the communication interface of the interface unit 22, and may also be input to the input devices of the interface unit 22.

[0051] The interface unit 22 outputs, for example, various types of information. The interface unit 22 includes, for example, a display device such as a CRT display, a liquid crystal display, or an organic EL display, and a speaker. The interface unit 22 may be configured as an interface that connects these display devices or speakers to the estimation device 2. Therefore, the interface unit 22 may output, for example, information input to an input device of the interface unit 22 as an image or sound.

[0052] The storage unit 23 is configured using a computer-readable storage medium device (non-transitory computer-readable recording medium) such as a magnetic hard disk device or a semiconductor storage device. The storage unit 23 stores various information related to the estimation device 2. The storage unit 23 stores various information generated by the operation of the control unit 21, for example. The storage unit 23 may exist on a cloud, for example.

[0053] 8 is a flowchart showing an example of the flow of processing executed by the estimation device 2 in the embodiment. The control unit 21 executes the estimation processing (step S201).

[0054] The learning device 1 configured in this manner executes the learning process, which reduces the burden of determining the cost of each position in a path planning problem, as described in <Effects of the Learning Process>.

[0055] Furthermore, the estimation device 2 configured in this manner performs estimation using the learning results obtained by the learning device 1. Therefore, the estimation device 2 can reduce the burden required to determine the cost of each position in a path planning problem.

[0056] Furthermore, the information processing system 100 configured in this manner includes the learning device 1. Therefore, the information processing system 100 can reduce the burden required to determine the cost of each position in a path planning problem.

[0057] (Variation) The learning device 1 may be implemented using a plurality of information processing devices connected to each other via a network so that they can communicate with each other. In this case, the processes executed by the control unit 11 may be distributed among the plurality of information processing devices.

[0058] The estimation device 2 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, the processes executed by the control unit 21 may be distributed among the plurality of information processing devices.

[0059] All or part of the functions of the information processing system 100, the learning device 1, and the estimation device 2 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may be transmitted via a telecommunications line.

[0060] The control unit 21 is an example of an estimation unit.

[0061] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention. [Explanation of symbols]

[0062] 100...information processing system, 1...learning device, 2...estimation device, 11...control unit, 12...interface unit, 13...storage unit, 21...control unit, 22...interface unit, 23...storage unit, 91...processor, 92...memory, 93...processor, 94...memory

Claims

1. a control unit that performs learning of a learning object model, which is a mathematical model of a learning object; Equipped with The learning target model is a mathematical model that estimates a cost of each position in a two- or three-dimensional space in a path planning problem that obtains a path of movement of a moving object in the space that satisfies the conditions as an optimal solution, based on first input information that indicates information about each position in the space and second input information that is text data in natural language indicating conditions that a moving object moving in the space must satisfy. Learning device.

2. The learning is supervised learning, In the supervised learning, data to which a cost is assigned as a correct answer label for each pair of the first input information and the second input information is used as training data. The learning device according to claim 1 .

3. The training model obtains an embedding vector of the text data based on the second input information, and estimates the cost using the obtained embedding vector. The learning device according to claim 1 .

4. the first input information is a result of a predetermined embedding process performed on the information indicating the space; The learning device according to claim 1 .

5. The information about each location indicated by the first input information indicates a classification to which each location belongs. The learning device according to claim 1 .

6. a control unit that learns a learning object model, which is a mathematical model of a learning object, and the learning object model is a mathematical model that estimates, based on first input information that indicates information about each position in a two- or three-dimensional space and second input information that is text data in a natural language indicating conditions that a movement path of a moving object moving through the space must satisfy, a cost of each position in the two- or three-dimensional space in a path planning problem that obtains a movement path within the space of the moving object as an optimal solution, the movement path satisfying the conditions; An estimation device comprising:

7. a control step in which the computer learns a learning target model, which is a mathematical model of the learning target; and The learning target model is a mathematical model that estimates a cost of each position in a two- or three-dimensional space in a path planning problem that obtains a path of movement of a moving object in the space that satisfies the conditions as an optimal solution, based on first input information that indicates information about each position in the space and second input information that is text data in natural language indicating conditions that a moving object moving in the space must satisfy. How to learn.

8. a control step in which a computer learns a learning object model, which is a mathematical model to be learned, wherein the learning object model is a mathematical model that estimates, based on first input information indicating information about each position in a two- or three-dimensional space and second input information being text data indicating in natural language conditions to be satisfied by a movement path of a moving object moving through the space, a cost of each position in the space in a path planning problem that obtains a movement path of the moving object within the space as an optimal solution, the movement path satisfying the conditions; and an estimation step in which a computer performs estimation using the learned learning object model obtained by the learning method. An estimation method having:

9. A program for causing a computer to function as the learning device according to any one of claims 1 to 5.

10. A program for causing a computer to function as the estimation device according to claim 6.

Images